This document provides a project report on building a descriptor-based support vector machine (SVM) for document categorization. It introduces SVMs and discusses how they were implemented for this project, including transforming data, scaling, using an RBF kernel, and training and assigning documents. The architecture of the SVM-based system is described, including training SVMs on descriptors and assigning descriptors to new documents. Experiments were conducted on a testbed using 5 descriptors, and recall, precision, and correct rate metrics were used to evaluate the results. In conclusion, the document demonstrates applying SVMs to automatically categorize documents based on their descriptors.